News Article: "Fun With Numbers" - Advanced Stats Talk Here

[StefanW]

Not sure if this has been mentioned but the only way to truly advance the stats is to identify what you want and then find a way to collect the data that can give you what you want in every rink. If we only have the data we have now, then the stats revolution is over and we have what we have.

I guess the psychology revolution is over because they have what they have. Ditto stats in sociology, criminology, economics, etc etc etc.

If you follow the links I posted above you'll see two brand new sites that organize data in ways that have not been done up until now. Add to that the upcoming trial of SportsVu, and you will see that things are far from stagnant. In fact, the stats revolution in hockey is accelerating like crazy at the moment.

 

[StefanW]

Interesting read over at Senstats; http://senstats.com/a-season-review/

Very data heavy, and most of it not the variety you see plastered across every other blog so it's a bit of a fresh look at things sens from an Analytics perspective.

It was predictably not kind to the GSN line, Kassian, or Spezza's defensive work.

Now the weekend is here I finally had the time to pour through this article. It is a great read, with a lot of very good insights. I personally enjoyed seeing the correlation matrix between metrics the most--it is gold. I think the quick summary of our up and coming players as "ready out of the box depth" is bang on.

Really a lot of info in there too that we already know from watching last season, but with numbers behind it. E.g. it is no shock that Spezza struggled on D, but I guess it had to be said.

 

[BigRig4]

Anyone else notice how great Boro is at preventing shots against? Goes to show that he's a black hole offensively to still have such bad corsi I guess.

 

[DrEasy]

I think he's saying that the people that give out the awards are cave bros

Thanks, that's exactly what I was trying to say. Advanced stats hate Jeter's defense, yet a jury of peers rewarded him year after year with the highest award for defense. It doesn't even matter whether Jeter was indeed great or not defensively. All this proves is that the people giving him the award did not care about what the advanced stats said. If it's a similar thing with the Jack Adams, we can't draw any conclusion from the award regarding McStashe's use of advanced stats.

Kassian played 5 mins a night, and his starts became less frequent as the season went on. His use had very little to do with whether or not Ottawa was paying attention to the stats. He was there for intimidation, and his mins were managed knowing full well his limitations (ie not being all that good at the sport of hockey, relative to his peers that is).

Truth is, Kassian and GSN got the mins they did in part because we lacked depth. We didn't have the players to put together a 4th line the coach could trust, and guys like DaCosta, Conacher, and Pageau we not getting it done.

Worse than the 5 minutes a game that Kassian "played", it's the fact that he occupied a roster spot at all. Let's not forget all those games that he was a healthy scratch! You can bring up a guy like Da Costa (who *was* getting it done) and you give him more minutes, and you decrease the minutes given to GN (Smith is fine indeed) or even just scratch Neil. Then you can use Neil purely for intimidation. Or you can bring up Boro and you play him over Phillips when you want the added fear factor. I'm pretty sure Boro is more intimidating on the ice than Kassian. And he's much more useful. And that's not even questioning whether you need that intimidation factor at all.

But I don't want to derail the conversation. All I'm saying is that I see no evidence that our coaching staff cares about advanced stats. And, who knows, maybe they're right. Hockey stats are still in their infancy after all, and there's plenty of room for new findings that might contradict what we think today.

 

[Benttheknee]

That's what I find interesting about the senstats site. He manually tracked zone entries and exits for all last years sens games; it's labour intensive, but unique from a lot of the other stuff out there and too me is a lot more informative than just Corsi for example.

If they really want to take this to the next level, they need to install a method to track the puck and players while on the ice.

 

[Micklebot]

If they really want to take this to the next level, they need to install a method to track the puck and players while on the ice.

Sportvu is probably coming to the NHL

 

[Micklebot]

Thanks, that's exactly what I was trying to say. Advanced stats hate Jeter's defense, yet a jury of peers rewarded him year after year with the highest award for defense. It doesn't even matter whether Jeter was indeed great or not defensively. All this proves is that the people giving him the award did not care about what the advanced stats said. If it's a similar thing with the Jack Adams, we can't draw any conclusion from the award regarding McStashe's use of advanced stats.

Coaching is purely results driven; wins and loses. Nobody cares how you do it. Defensive play in baseball and winning the golden glove are a little different in that it's much more difficult to evaluate, and reputation bias comes into play.

When McLean won the Jack Adams, he didn't have much in the way of reputation, and the teams advanced stats were actually pretty good. They also had lots of additional factors (mostly injuries) that made the feat more impressive.

While there was certainly room for criticism of McLean last year, his Jack Adams year was pretty solid.

Worse than the 5 minutes a game that Kassian "played", it's the fact that he occupied a roster spot at all. Let's not forget all those games that he was a healthy scratch! You can bring up a guy like Da Costa (who *was* getting it done) and you give him more minutes, and you decrease the minutes given to GN (Smith is fine indeed) or even just scratch Neil. Then you can use Neil purely for intimidation. Or you can bring up Boro and you play him over Phillips when you want the added fear factor. I'm pretty sure Boro is more intimidating on the ice than Kassian. And he's much more useful. And that's not even questioning whether you need that intimidation factor at all.

But I don't want to derail the conversation. All I'm saying is that I see no evidence that our coaching staff cares about advanced stats. And, who knows, maybe they're right. Hockey stats are still in their infancy after all, and there's plenty of room for new findings that might contradict what we think today.

If Neil was still intimidating, we never would have traded for Kassian. Boro doesn't intimidate either, and really isn't much of a fighter. He'll drop the gloves, but that's not really why you have him out there.

Kassian being on the roster doesn't prevent guys like DaCosta from playing, he's inserted into the lineup selectively, and serves a far different role. How is Kassian being a healthy scratch worse than DaCosta being a healthy scratch? DaCosta rarely got any minutes when he was in the lineup, not even breaking 10 mins in half his games, none of that is Kassian's fault. If we had more depth on the wings last year, maybe DaCosta would have gotten more shifts, and played more, but we lacked the depth to make an effective 4th line. Shortening the bench isn't feasible all season long, but worked fine in Kassian's limited appearances.

Kassian didn't hurt player development; not having enough solid players to trust the 4th line with mins hurt development.

 

[StefanW]

The bias with Jack Adams is partly about reputation, but also has a lot to do with a "getting results with less" scenarios. MacLeans COTY win, and when he was a finalist, occurred in years when the Sens were expected to miss the playoffs. Those teams went well past pre-season expectations. I don't think it had anything to do with improving underlying numbers.

GM of the year is probably the worst NHL award with respect to reputation bias. As far as I know, none of them really use analytics in any meaningful way.

 

[Micklebot]

The bias with Jack Adams is partly about reputation, but also has a lot to do with a "getting results with less" scenarios. MacLeans COTY win, and when he was a finalist, occurred in years when the Sens were expected to miss the playoffs. Those teams went well past pre-season expectations. I don't think it had anything to do with improving underlying numbers.

GM of the year is probably the worst NHL award with respect to reputation bias. As far as I know, none of them really use analytics in any meaningful way.

This is true. I actually misread the post I was responding to, he is right that winning coty gives no indication as to whether or not the coach is aware of or adjusts his tactics based on analytics.

 

[Busboy]

The bias with Jack Adams is partly about reputation, but also has a lot to do with a "getting results with less" scenarios. MacLeans COTY win, and when he was a finalist, occurred in years when the Sens were expected to miss the playoffs. Those teams went well past pre-season expectations. I don't think it had anything to do with improving underlying numbers.

GM of the year is probably the worst NHL award with respect to reputation bias. As far as I know, none of them really use analytics in any meaningful way.

I didn't even know they awarded GM of the year. It seems like a tough one to judge, maybe that's why it doesn't get enough attention for me to know about it.

 

[Micklebot]

Another article looking at zone entry data; he combines a metric he developed using zone entry data with Corsi and got a pretty strong correlation with RegWin%

http://hockeyanalysis.com/2014/08/26/team-zone-entry-data-predicting-standings/

Of note, Ottawa and NJD were the only non-playoff teams with a positive percentage in his homebrew corsi/ZoneEntry fancy stat.

 

[Caeldan]

Another article looking at zone entry data; he combines a metric he developed using zone entry data with Corsi and got a pretty strong correlation with RegWin%

http://hockeyanalysis.com/2014/08/26/team-zone-entry-data-predicting-standings/

Of note, Ottawa and NJD were the only non-playoff teams with a positive percentage in his homebrew corsi/ZoneEntry fancy stat.

I dislike the dump in being treated as neutral...
Should have taken it a step further if they're doing carry in vs failed carry in and have a dump in win vs dump in loss (ie a dump in win is one where you dump it in, then regain possession on entry)

But I do think it makes sense that it's marginally better than CF% since basically it's a more manual count of possession if you are carrying the puck in. The entire purpose of CF is to try to guesstimate possession from box scores since there's nothing better tracked automatically.

 

[God Says No]

Another article looking at zone entry data; he combines a metric he developed using zone entry data with Corsi and got a pretty strong correlation with RegWin%

http://hockeyanalysis.com/2014/08/26/team-zone-entry-data-predicting-standings/

Of note, Ottawa and NJD were the only non-playoff teams with a positive percentage in his homebrew corsi/ZoneEntry fancy stat.

LOL at montreal being one of the bottom teams in the table who made the playoffs. If it wasn't for Price they would be in real trouble.

 

[Micklebot]

I dislike the dump in being treated as neutral...
Should have taken it a step further if they're doing carry in vs failed carry in and have a dump in win vs dump in loss (ie a dump in win is one where you dump it in, then regain possession on entry)

But I do think it makes sense that it's marginally better than CF% since basically it's a more manual count of possession if you are carrying the puck in. The entire purpose of CF is to try to guesstimate possession from box scores since there's nothing better tracked automatically.

I think he's using zone entry data as a proxy for shot quality, the premise being that corsi events resulting form a carry in are more valuable than corsi events for a dump in. He's also presuming that a failed carry in is equally likely to result in a high quality chance against.

So if his assumptions are correct, it's not so much that the zone entry data does a better job of approximating OZ possession, but rather that he's adding a qualitative component to the Corsi metric.

 

[Benttheknee]

Sportvu is probably coming to the NHL

I read that shortly after making my post. It will be interesting to see how this changes the game and the way players are compensated. I bet Eric Condra is drooling

 

[StefanW]

I read that shortly after making my post. It will be interesting to see how this changes the game and the way players are compensated. I bet Eric Condra is drooling

I suspect teams will continue to pay a premium for goals. Condra has outstanding underlying numbers, but he can't score to save his life. That makes him an awesome bottom 6 guy, providing he is on a reasonable contract. Nothing SportVu will pull out is going to change any of this.

I'm actually the most excited about player speed and acceleration metrics, providing the become available.

 

[18Hossa]

New stas article on the Sens' site http://senators.nhl.com/club/m_blogs.htm?id=31314

 

[Do Make Say Think]

They really need to put a better picture of MacArthur on the website

That dear in the headlights look he has is disconcerting

 

[Caeldan]

Another on the subject of Corsi...
It's an older article - was looking for some information on influences, found this kinda interesting.
Basically saying that FO% has the highest impact on a CF%, followed by ZS%... and that you could calculate a player's CF% to about +/-7 points.

So then they took actual CF%, subtracted out the predicted - this is what's left (so I believe it is the author's theory that this is the part that the player actually influences)... I pulled out the Senators players from the list (570 players) and it looks like the following:

Player|Pos.|FO%|ZS%|CF%|PredCF%|AdjCF%
Clarke MacArthur|L|50.10%|49.10%|54.10%|49.58%|4.52%
Kyle Turris|C|49.20%|47.40%|52.80%|48.62%|4.18%
Erik Karlsson|D|50.70%|55.40%|54.80%|51.99%|2.81%
Chris Phillips|D|52.20%|46.60%|51.70%|49.58%|2.12%
Zack Smith|C|52.70%|41.80%|50.20%|48.13%|2.07%
Eric Gryba|D|53.70%|43.20%|50.90%|49.02%|1.88%
Colin Greening|C|51.40%|46.20%|50.90%|49.11%|1.79%
Chris Neil|R|52.00%|45.40%|50.70%|49.08%|1.62%
Bobby Ryan|R|47.80%|51.60%|51.10%|49.49%|1.61%
Mika Zibanejad|C|51.20%|56.50%|53.80%|52.58%|1.22%
Marc Methot|D|51.60%|52.10%|52.30%|51.23%|1.07%
Patrick Wiercioch|D|51.90%|54.40%|53.10%|52.14%|0.96%
Erik Condra|R|54.30%|53.80%|53.70%|52.92%|0.78%
Jared Cowen|D|50.80%|50.30%|50.90%|50.28%|0.62%
Cody Ceci|D|51.10%|52.50%|51.00%|51.16%|-0.16%
Ales Hemsky|R|52.30%|47.60%|49.60%|49.96%|-0.36%
Jason Spezza|C|53.90%|54.90%|52.30%|53.14%|-0.84%
Milan Michalek|L|54.00%|57.00%|51.90%|53.90%|-2.00%

Not sure if there's anything really to pull out of that - but if you want to read it's here:
http://www.boysonthebus.com/2014/05/05/the-influence-of-face-offs-and-zone-starts-on-corsi/

 

[operasen]

Anyone hear the TSN interview with Mirtle (Globe and Mail) on Advanced stats and how its blowing up all over?

He indicated the new site he liked for this, since Kyle Dubas and extraskater.com are now off the market.

I looked at waronice.com. Pretty interesting I must say.

 

[Upgrayedd]

Interesting hearing on the radio yesterday that Ottawa based company PowerScout Hockey is currently pitching their optical tracking technology to the NHL.

A little further investigation reveals a pretty cool technology!

 

[operasen]

Interesting hearing on the radio yesterday that Ottawa based company PowerScout Hockey is currently pitching their optical tracking technology to the NHL.

A little further investigation reveals a pretty cool technology!

Heard that as well. Sounds like it would create an entirely new department of statistics on some teams.

Ottawa based. Props to that. http://www.powerscouthockey.com/

The MoneyBall of NHL Hockey

 

[StefanW]

Interesting hearing on the radio yesterday that Ottawa based company PowerScout Hockey is currently pitching their optical tracking technology to the NHL.

A little further investigation reveals a pretty cool technology!

I'm wondering how many teams with go the external consulting route rather than forming their own internal analytics department.

 

[Upgrayedd]

I'm wondering how many teams with go the external consulting route rather than forming their own internal analytics department.

Hmmm thats an interesting one to ponder...there are seemingly benefits to both routes. The portion of said interview that piqued my interest was the part about teams doing this on the down low so as to hold a competitive advantage, wonder if and how long thats been going on and by who? Got me thinking of a scenario where say a team had perhaps gotten wind of this company Power Scout before the advanced stat trend hit mainstream, liked what they saw and were able to use the data and bought the technology exclusively. ..what that would set in motion in terms of the league overall. I am of course not sure of the legalities regarding this scenario in terms of NHL policies.

 

[StefanW]

Hmmm thats an interesting one to ponder...there are seemingly benefits to both routes. The portion of said interview that piqued my interest was the part about teams doing this on the down low so as to hold a competitive advantage, wonder if and how long thats been going on and by who? Got me thinking of a scenario where say a team had perhaps gotten wind of this company Power Scout before the advanced stat trend hit mainstream, liked what they saw and were able to use the data and bought the technology exclusively. ..what that would set in motion in terms of the league overall. I am of course not sure of the legalities regarding this scenario in terms of NHL policies.

Advanced stats are only an advantage to a club if they find an angle that other teams have not considered. If they are doing everything the same way as everyone else the game is about not falling behind instead of getting ahead. That is the big push toward innovation and development of new metrics.

I would caution against accepting teams at face value when they say they are already doing advanced stats and it is hush-hush. There is a discussion of this earlier in the thread.